图像分类model显示图执行错误

Question

我正在进行一项多类 covid-19 胸部 X 光检查挑战，在建模的最后一部分遇到了问题。 它一直显示“图形执行错误”。 似乎是图像大小问题，但我不知道如何解决。 我需要将给定的数据分为三个不同的类别：“Nofinding”、“covid19”和“thoraxdisease”，我认为使用分类 class 是正确的，而不是使用二进制。 但

**项目链接在这里： https://www.kaggle.com/competitions/csc532-2/data **这是我的代码：

    train_path = '/content/drive/MyDrive/covid19/csc532-2/DLAI3_Phase3/DLAI3_Phase3'

train_COVID19_1 = glob.glob(train_path+"/COVID-19/*.png")
train_NOFINDING_1 = glob.glob(train_path+"/NOFINDING/*.png")
train_THORAXDISEASE_1 = glob.glob(train_path+"/THORAXDISEASE/*.png")

train_COVID19_2 = glob.glob(train_path+"/COVID-19/*.jpg")
train_NOFINDING_2 = glob.glob(train_path+"/NOFINDING/*.jpg")
train_THORAXDISEASE_2 = glob.glob(train_path+"/THORAXDISEASE/*.jpg")

train_COVID19_3 = glob.glob(train_path+"/COVID-19/*.jpeg")
train_NOFINDING_3 = glob.glob(train_path+"/NOFINDING/*.jpeg")
train_THORAXDISEASE_3 = glob.glob(train_path+"/THORAXDISEASE/*.jpeg")

    train_list = [x for x in train_COVID19_1]
train_list.extend([x for x in train_COVID19_2])
train_list.extend([x for x in train_COVID19_3])
train_list.extend([x for x in train_NOFINDING_1])
train_list.extend([x for x in train_NOFINDING_2])
train_list.extend([x for x in train_NOFINDING_3])
train_list.extend([x for x in train_THORAXDISEASE_1])
train_list.extend([x for x in train_THORAXDISEASE_2])
train_list.extend([x for x in train_THORAXDISEASE_3])

df_train = pd.DataFrame(np.concatenate([['COVID19']*(len(train_COVID19_1)+len(train_COVID19_2)+len(train_COVID19_3)), 
                                        ['NOFINDING']*(len(train_NOFINDING_1)+len(train_NOFINDING_2)+len(train_NOFINDING_3)),
                                        ['THORAXDISEASE']*(len(train_THORAXDISEASE_1)+len(train_THORAXDISEASE_2)+len(train_THORAXDISEASE_3))]), 
                        columns = ['class'])
df_train['image'] = [x for x in train_list]

df_train

    validate_path = '/content/drive/MyDrive/covid19/csc532-2/VALIDATE'

validate_1 = glob.glob(validate_path+"/VALIDATE/*.png")

validate_list = [x for x in validate_1]
             
df_validate = pd.DataFrame(np.concatenate([['Validate']*len(validate_1)]), columns = ['class'])
df_validate['image'] = [x for x in validate_list]

df_validate

    from tensorflow.random import set_seed
set_seed(42)

    IMG_SIZE = 224
BATCH = 32
SEED = 42

    from sklearn.model_selection import train_test_split
from numpy.random import seed
train_df, test_df = train_test_split(df_train, test_size = 0.20, random_state = SEED, stratify = df_train['class'])

    train_df

    test_df

    # https://vijayabhaskar96.medium.com/tutorial-on-keras-flow-from-dataframe-1fd4493d237c

train_datagen = ImageDataGenerator(rescale=1/255.,
                                  zoom_range = 0.1,
                                  #rotation_range = 0.1,
                                  width_shift_range = 0.1,
                                  height_shift_range = 0.1)

val_datagen = ImageDataGenerator(rescale=1/255.)

ds_train = train_datagen.flow_from_dataframe(train_df,
                                             #directory=train_path, #dataframe contains the full paths
                                             x_col = 'image',
                                             y_col = 'class',
                                             target_size = (IMG_SIZE, IMG_SIZE),
                                             class_mode = 'categorical',
                                             batch_size = BATCH,
                                             seed = SEED)

ds_val = val_datagen.flow_from_dataframe(test_df,
                                            #directory=train_path,
                                            x_col = 'image',
                                            y_col = 'class',
                                            target_size = (IMG_SIZE, IMG_SIZE),
                                            class_mode = 'categorical',
                                            batch_size = BATCH,
                                            seed = SEED)

ds_test = val_datagen.flow_from_dataframe(df_validate,
                                            #directory=test_path,
                                            x_col = 'image',
                                            y_col = 'class',
                                            target_size = (IMG_SIZE, IMG_SIZE),
                                            class_mode = 'categorical',
                                            batch_size = 1,
                                            shuffle = False)

    keras.backend.clear_session()

model = get_model()
model.compile(loss='binary_crossentropy'
              , optimizer = keras.optimizers.Adam(learning_rate=3e-5), metrics='binary_accuracy')
 
model.summary()

最后，这是当我在上面将 binary_crossentropy 切换为 categorical_crossentropy 时出现错误的部分。

    history = model_pretrained.fit(ds_train,
          batch_size = BATCH, epochs = 30,
          validation_data=ds_val,
          callbacks=[early_stopping, plateau],
          steps_per_epoch=(len(train_df)/BATCH),
          validation_steps=(len(test_df)/BATCH));

我在想这可能是图像大小不匹配的问题，但我已经将它们设置为 244x244，如上所示。

我刚刚改变

Answer 1

你能不能model = get_model()也为 model 附上这个 function。

与flow_from_dataframe一样，这里有flow_from_directory直接从目录加载数据的方法。

img_datagen = tf.keras.preprocessing.image.ImageDataGenerator(
    rescale=1.0/255.0,
    validation_split=0.2,
    rotation_range=5,
    shear_range=0.3,
    zoom_range=0.3,
    width_shift_range=0.05,
    height_shift_range=0.05,
    horizontal_flip=True,
    vertical_flip=True,
)

train_ds = img_datagen.flow_from_directory('path',subset="training", class_mode = 'categorical', batch_size=16, target_size=(224,224))
val_ds = img_datagen.flow_from_directory('path',subset="validation", class_mode = 'categorical', batch_size=16, target_size=(224,224))

model.compile(loss='categorical_crossentropy', optimizer = keras.optimizers.Adam(learning_rate=0.001), metrics='accuracy')
 
model.summary()